© 2019 Hey Machine Learning

  • Maryna Hlaiboroda

5 Companies that Use Machine Learning.Part 2

Artificial intelligence is now touching all industries, from healthcare and advertising to entertainment and tech. AI commerce companies have raised more $1B in 201 deals since 2012. Around 42% of these deals have gone to commerce startups outside the United States. Startups in France accounted for 6.5% of the deals share, followed by Canada (5.5%) and India (5.5%).

We continue to compile a list of companies from various fields that use Artificial Intelligence and Machine Learning to increase the efficiency of their business.


Opera Software. Personalization of news feed



Opera Software is a Norwegian software company, primarily known for its desktop Opera web browser, and mobile web browsers Opera Mini and Opera Mobile. Their total market share in April 2018 was 3.83%, the fifth largest in the world. Therefore, the software manufacturer is actively looking for new ways to attract more users.

On December 19, 2017, Opera developers increased the number of Newsreader news service users from 2 million to 100 million people using Machine Learning algorithms. The average user Opera spends 40 minutes every day inside the Newsreader and looking through between 65 and 81 news stories per day. But if users without any effort will get access to interesting news for them, then the popularity of news aggregators will increase dramatically. This happened due to ML.

In addition, in the nearest future the company plans to launch the news program Opera Habari, which will include elements of Artificial Intelligence. The task of AI will be to study the preferences of users and provide news that will be of interest to a particular reader. Opera Software undertakes to release the application "soon", but doesn't specify when.


Let's Enhance and improve the quality of photos



Neural Networks works with images well, this is a well- known fact. One of the proofs is the online service letsenhance.io of the Ukrainian start-up Let's Enhance. The project allows you to improve the detail of the photo and increase its resolution 4 times. The creators used cutting-edge Image Super Resolution technology based of Machine Learning when developing the service. A Neural Network, trained on a huge base of real photographs, has learned to restore details and contours, based on previously acquired knowledge of typical objects and textures from the real world.

Photo processing takes place using Deep Convolutional Neural Networks, on the basis of which there are 4 tools:

  • Boring filter eliminates JPEG artifacts, and preserves colors, clarity of lines and structures

  • Magic filter hallucinates additional details initially missing in the image

  • Tone improves texture and shades

  • Color improves color and appearance

If you want to use the service, you need to register. For free users Let's Enhance creators set the limit to 5 Megapixels and 5 Megabytes for every photo uploaded. A paid subscription costs $7 per month and removes all limits in the size and number of files.


Adobe Systems. Editing photos and videos




The American developer of software Adobe Systems at the yearly conference Adobe Max tells about the technologies that will someday enter into Photoshop, Premiere and Lightroom.

In 2017, Adobe released an update to the photo editor Lightroom in which developers improved the functionality of automatic photo processing introducing algorithms of Artificial Intelligence. Automatic mode compares the image with a large number of previously processed files by professionals, chooses the optimal settings, picks the appropriate filters and applies effects. Using the algorithms of Machine Learning the photo editor takes considered decisions about the methods of image correction.

Also in 2017, the corporation introduced many new products based on the Adobe Sensei AI Machine Learning technology. The most remarkable of them are:

  • Scene Stitch If earlier Photoshop was looking for colors and textures inside the image itself, now the system analyzes the whole database of pictures and prompts which ones are best for editing this photo.

  • Cloak This function tracks an unnecessary object on the video and replaces it using a method like Scene Stitch.

  • Project Puppetron Although the filters have never been the strength of Photoshop, now the demonstrated Adobe technology simulates the art style much more accurately than the Prisma application.

  • Project Scribbler This AI-tool paints black and white images using different textures and colors.

  • SonicScape This function allows you to adjust the sound for VR-content: when the user turns his head changes not only the panorama, but also the sound.

  • PhysicsPak The tool allows you to put a lot of objects in a certain shape. For example, draw the letter "A" from the silhouettes of cats.

In 2018, Adobe also provided an updated version based on Adobe Sensei AI is the Select Subject tool. It recognizes the areas required for selection in the photo with high accuracy.


MindGeek. Thematic tag porn




On October 4, 2017, pornosite PornHub, owned by the global IT company MindGeek, has announced it is launching a new AI-powered model that uses computer vision technology to autonomously detect and identify adult performers by their name.

Engineers coached the AI-model on thousands of videos along with official photos of performers to learn how to return names accurately. To limit the margin of error, the adult website is asking users to validate whether the tags and labels provided by AI model are needed in correction. To do so, users can either upvote or downvote the accompanying descriptions, depending on their accuracy. This is how the algorithm gets smarter.

In addition to identifying performers, the algorithm have the capacity to distinguish between different categories. It will for instance mark videos shot in outdoor settings under the 'Public' category and clips featuring blonde performers in the 'Blonde' category.

The PornHub plans to run its entire five-million-strong catalog through the AI model over the 2018 year in hopes to make it easier for users to find the clips that most closely align with their desires. Also, according to representatives of video hosting, the site already uses the ContentID algorithm similar to what works on YouTube. It automatically blocks videos that violate copyrights.


Nasdaq. Protection against fraud



After a series of financial scandals, the American stock exchange Nasdaq applied artificial intelligence algorithms to protect against fraud. Using the platform of cognitive calculations developed by Digital Reasoning, Nasdaq analyzes the communication models of traders to understand their hidden intentions and actions, and also reveals fraudulent schemes.

Prior to using the Digital Reasoning technology, Nasdaq had evaluated products for about a year across all the major providers of electronic communication monitoring technology. It determined that the legacy keyword/lexicon-based products were not effective in identifying communications that indicated collusive or manipulative intent and generated excessive false positive alerts, said Michael O'Brien, head of product management, risk, and surveillance solutions at Nasdaq.

The exchange is using the Digital Reasoning platform in conjunction with SMARTS, Nasdaq's own AI-based surveillance technology. Nasdaq and Digital Reasoning have a proof-of-concept project underway with a joint bank customer, in which channels have been set up for the passing of information between the two systems. In the first phase of the project communications identified by Digital Reasoning as containing collusive, manipulative, or other anomalous language are being tagged to related trade surveillance alerts in the SMARTS interface.

"This represents a very significant step forward in trade monitoring, as it enables the analyst to have a view of the anomalous trading and the intent of the traders involved in that trading," O'Brien said.